Antibody human adaptation is a generic term describing the engineering of xenogeneic monoclonal antibodies (mAbs) against human therapeutic targets to maximally replace the xenogeneic sequences with human antibody sequences while preserving their antigen-binding specificities. The aim is to reduce the immunogenicity of these antibodies to improve their therapeutic properties. The engineered antibodies generated are also known in the art as humanized or CDR-grafted antibodies.
Currently, the most widely used technique for antibody human adaptation is known as “CDR grafting.” The scientific basis of this technology is that the binding specificity of an antibody resides primarily within the three hypervariable loops known as the complementarity determining regions (CDRs) of its light and heavy chain variable regions (V-regions), whereas the more conserved framework regions (framework, FW; framework region, FR) provide structure support function. By grafting the CDRs to an appropriately selected FW, some or all of the antibody-binding activity can be transferred to the resulting recombinant antibody. The first demonstration of the transfer of specificity by CDR grafting was for a hapten nitrophenol (Jones et al., Nature 321:522-525 (1986)).
Since the methodology for defining CDRs has been well established, the key to CDR grafting is the selection of a most appropriate human antibody acceptor for the graft. Various strategies have been developed to select human antibody acceptors with the highest similarities to the amino acid sequences of donor CDRs or donor FW, or to the donor structures. All these “best fit” strategies, while appearing very rational, are in fact based on one assumption, i.e., a resulting recombinant antibody that is most similar (in amino acid sequence or in structure) to the original antibody will best preserve the original antigen binding activity. While these strategies have all been successfully applied to generate therapeutic antibodies (e.g., Tempest et al., Biotechnology 9:266-71 (1991), Gorman et al., Proc. Natl. Acad. Sci. USA 88:4181-4185 (1991), Co et al., J. Immunol. 152:2968-76 (1994)), the underlying hypothesis has never been seriously tested.
One potential problem of the best-fit strategies is that the criteria of best fits are mathematical, but not necessarily biological. The fitness measured by the degree of homology, for example, is the sum of numerical values assigned to identical, homologous, and dissimilar amino acid residues or nucleic acid sequences. Although these assigned values have largely been validated in many other homology evaluating systems, the fine differences that may not be significant for other systems could be important for calculating the best fits in antibody human adaptation.
A related problem is, given two acceptors with identical or very close degree of total fitness for the donor, their local fitness in different FRs may be different. In short, a mathematic model has not yet been validated to satisfy the requirement of calculating the best fits in donor-acceptor relationship in antibody engineering.
A further complication relates to the interactions between the two chains of an antibody: a best-fit heavy chain acceptor and a best-fit light chain acceptor may not fit with each other to best conserve the binding activity of the donor. No tool is available to evaluate interchain fitness. Investigators have paired heavy and light chains of several antibodies against the same epitope to try to find a better pairing. However, this has not been attempted in antibody human adaptation.
In theory, all human germline sequences have been sequenced and are available for antibody FW searching. In practice, however, the majority of human V regions that have been used so far in antibody humanization are from mature antibody genes, often those of myeloma proteins. They are likely to contain somatic mutations. These mutations are unique to the individual from which the rearranged genes were derived, and hence will be seen as foreign by other individuals. Germline database sequences are generally more suitable for antibody humanization from this perspective. However, there are only a few dozen V genes and fewer J genes available. Hence, in some instances, it may be difficult to find a germline framework that is highly compatible with the non-human sequence. In contrast, the number of available mature human antibody gene sequences is a few orders of magnitude higher than the human germline sequences. Thus, it is highly likely that one can obtain a compatible sequence in this larger dataset.
A problem of using mature antibody genes for acceptor FW is that not all of the potential V-J combinations for light chain or V-D-J combinations for heavy chain are represented in the mature genes. Thus, situations can arise in which a closely matching V gene is linked to a poorly matching J segment. The humanization of the mouse anti-Tac monoclonal antibody described by Queen et al., (Proc. Natl. Acad. Sci. USA 86:10029-10033 (1989)) is an example. Comparison of the anti-Tac VH region to the NBRF-PIR database (http:_//www_psc_edu/general/software/packages/nbrf-pir/nbrf_html) indicated that the VH region of the human myeloma protein Eu had the highest degree of homology (57% identical over VDJH). However, framework 4 of the Eu VH region has several amino acids, presumably encoded by the Eu JH segment, that are atypical of human JH segments. This resulted in a poor match between the Eu framework 4 and that of anti-Tac. Separate comparison of the anti-Tac JH region (framework 4 and the framework 4—proximal end of CDR3) to the amino acid sequences of the known functional human JH segments (of which there are 6) indicates that human JH4 is a much better-match than the Eu JH. This example suggests that separate comparisons of V and J elements are more advantageous than comparison of the whole variable regions between rodent and human antibody sequences. Currently, a tool for this type of separate comparison is not readily available.
Not all amino acids in the CDRs are involved in antigen binding. Thus, it has been proposed that the grafting of only those residues that are critical in antigen-antibody interaction—the so-called specificity determining residues grafting (SDR-grafting)—will further increase the content of human antibody sequences in the resulting recombinant antibody (Kashmiri et al., Methods 36:25-34 (2005); Gonzales et al., Mol Immunol. 40:337-49 (2004)). The application of this strategy requires information on the antibody structure as well as antibody-antigen contact residues, which are quite often unavailable. Even when such information is available, there is no systematic method to reliably identify the SDRs, and SDR-grafting remains so far mostly at the basic research level.
Recently, a novel strategy called “human framework shuffling” has been developed (Dall'Acqua et al., Methods 36:43-60 (2005)). This technique works by ligating DNA fragments encoding CDRs to DNA fragments encoding human FR1, FR2, FR3, and FR4, thus generating a library of all combinations between donor CDRs and human FRs. While this strategy has been successfully applied, there are two potential problems. First, the FRs of the resulting antibody, while all of human sources, are likely to be from non-contiguous FWs, and therefore unnatural. It remains to be seen whether these unnatural FWs will be immunogenic in humans. Second, the library, in theory, can be prohibitively large, and places a high demand on screening and assay resources.
Thus, a need exists for improved methods for making human-adapted antibodies.